MogileFS efficiency part two
dormando
dormando at rydia.net
Wed Jun 28 17:50:46 UTC 2006
Hey,
Another thing that's been brought up over here a few times...
The bulk of the time when we hit MogileFS for images we're loading 15+
smallish images on the same page (same as the LJ userpics). It would be
more efficient to have some method of indirectly or directly exposing
mogstored to the internet. When building the page output we can get the
paths for all of the images in one memcached call, then build out the
URLs with more direct (or encoded) information.
The goal is to avoid the:
perlbal -> webserver -> (memcached|tracker) -> perlbal
for every individual image load, and try to narrow it down to:
1) firewall -> perlbal -> mogstored
or better, but unlikely:
2) firewall -> mogstored
I'd be happy with #1
Every time I try to think about this I come up with tons of crap that
could break it. Don't want the URLs to change at every page load,
caching problems when paths change, have to patch mogstored to have a
second GET only port, etc. Nothing I'd be happy with. There should still
be *some* way that I'm missing to utilize the rendering page to make the
image display more lightweight?
-Dormando
More information about the mogilefs
mailing list